Handling Block-Missing Data with Model Combination Method
Here is my real example in paper, NYS School Performance.
Block-wise missing data is a common occurrence in regression and classification problem. In this paper, we consider one such situation, where one part of the observations is fully observed while the other part is only partially observed with some block of data missing. Although this type of problem has received some attention in the literature, we present a new model combination approach to effectively handle the block missing problem. The method works by using a linear combination of full and partial model. We propose several ways to estimate the combination parameter, which is shown to be consistent. We also show that the combination method works better than using the full or partial model alone. Simulation studies are conducted to evaluate our new method. The method is also applied to real datasets. One is about factors affecting performance to public schools in standardized tests. Another is about predicting accuracy to abdominal pain diagnoses in Emergency Department.